LaMAria
收藏资源简介:
LaMAria是一个用于评估多传感器视觉惯性定位与建图(VIO/SLAM)的基准数据集,它记录了在城市环境中使用Project Aria设备采集的大量轨迹数据。该数据集涵盖了独特的挑战,如低光照、曝光变化、移动平台、时变的校准等。数据集包含63个序列,每个序列平均覆盖1.5公里和26分钟,最长的一个达到2.87公里和48分钟。LaMAria数据集提供了基于稀疏控制点的厘米级精确位姿标注,使得能够评估极端轨迹,如夜间步行或乘车。此外,数据集还包含一个难度逐渐增加的测试集,以帮助深入分析和评估不成熟的VIO/SLAM方法。
LaMAria is a benchmark dataset for evaluating multi-sensor visual-inertial localization and mapping (VIO/SLAM). It documents a large corpus of trajectory data acquired using Project Aria devices in urban environments. This dataset encompasses unique challenges including low-light conditions, exposure variations, moving platforms, and time-varying calibrations. The dataset comprises 63 sequences, where each sequence covers an average of 1.5 kilometers and 26 minutes of recording, with the longest sequence spanning 2.87 kilometers and 48 minutes. LaMAria provides centimeter-level accurate pose annotations based on sparse control points, enabling the evaluation of extreme trajectories such as nighttime walking or vehicle rides. Additionally, the dataset includes a test set with gradually increasing difficulty to support in-depth analysis and evaluation of immature VIO/SLAM methods.

- 1Benchmarking Egocentric Visual-Inertial SLAM at City ScaleETH Zurich, Google, Meta Reality Labs Research, Microsoft Spatial AI Lab · 2025年



